Commenting on the Comment
Commenting on the Comment
复制标题
对评论进行评论
DOI：
--
复制
发表时间：
1994
期刊：
影响因子：
--
通讯作者：
J. Bader
中科院分区：
文献类型：
--
作者：
J. Bader
Dr. Bratthall's comment 1J Dent Res 73(2):497-498, 1994] regarding Dr. Douglass' editorial concerning evaluation of diagnostic tests in dentistry [J Dent Res 72(8):1170, 1993] presents an example that, in fact, exemplifies a common misconception of diagnostic test performance measures. The example describes a study wherein a test measuring oral hygiene and/or plaque is applied to three populations with markedly different caries incidence rates. Dr. Bratthall suggests that the test's performance measures of sensitivity and specificity will be different in the three populations. In fact, the sensitivity and specificity of the test, i.e., the ability to discriminate between those with and without some predetermined level of plaque, will remain unchanged in these populations. What will be different is the strength of the association between what is being tested (plaque level) and caries incidence. Thus, the "predictive power" of the test is being evaluated, but not its sensitivity and specificity to detect a specific marker, plaque level. The distinction is important, particularly because the same measures often are employed to evaluate a test's diagnostic accuracy and a "model's" predictive power. Clearly, while one doesn't use a plaque test to "diagnose" caries, such a test may be useful in predicting an individual's future caries experience. What Dr. Bratthall has described is actually a one-variable predictive model of the caries process. The model is obviously underspecified, which is not surprising considering that caries is a multifactorial disease. More generally, the performance of most predictive models will be reduced when they are applied in populations other than that from which the original data were collected. Thus, Dr. Bratthall's call for evaluation (validation) of predictive models in multiple populations is well-founded.
